Building a NHL Fantasy Predictor: How Much Playing Time is My Forward Getting Next Year?

Updated: January 10, 2018 at 7:09 pm by Ben Wendorf

Farting around with % of Attempted Shots (%AttSh) and coach data is fun, but the real bedrocks of fantasy prediction have to be line data and the focus of today’s post, playing time. The line data is useful for looking at playing time if you anticipate a drastic change for a forward, while the playing time data can be important in telling us something about how predictive it can be, and whether there are any trends within the data that we need to pay attention to. It’s one of the dirtier secrets of boxcar statistics (goals, assists, points), that they are frequently driven by playing time as much as they are driven by skill. Pierre Parenteau could’ve lived in AHL obscurity forever, but he received the ice-time opportunity that made him a great fantasy add the last couple of seasons.

Let’s have a look at what our ice-time data can tell us about the future…

A changing ice time landscape

The first thing you want to establish, before you jump to individual correlations of year-to-year 5v5 TOI/60 figures, is that 5v5 TOI for teams in general are pretty random. The correlation of Fenwick Close (considered one of the best indicators of team quality) to 5v5 TOI is 0.014 (n = 120), which essentially tells us it is random to team skill. The year-to-year correlation is high (0.75; correlation ranges from -1.0 to 1.0), but most of that is due to the fact that 5v5 TOI stays overwhelmingly in the range of 3,750 to 4,000 minutes per year, per team. So instead, I decided to see if 5v5 TOI revolved around a number, which is to ask at what amount of 5v5 TOI in one year does it become very likely that the team will see fewer 5v5 minutes the next year? 

Well then…first of all, we’re all over the map, but it’s hard not to notice that we’re staying out of the negative. Which means…

  • Avg Team 5v5 TOI 2007-08 … 3682.52 minutes
  • 2008-09 … 3739.61
  • 2009-10 … 3846.31
  • 2010-11 … 3897.34
  • 2011-12 … 3957.13

That can explain some of your decreased scoring right there. With that kind of trend, and that level of variability, I’m going to stick to projecting 4,000 even-strength minutes for each team (even though the statistical projection would be more like 4,030) and move on.

Progression with age

So, on to the forwards. It’s incredibly handy to us to know that, based on the last 5 years’ data for forwards (n = 1,475), the correlations of both 5v5 and 5v4 TOI/60 to the following seasons are 0.825 and 0.875, respectively. That’s all well and good, but it’d be nice to know where some of our variance comes from, even if it’s pretty minimal. Intuitively, you figure it has to be coming from age, as younger players tend to start out with less time, and when a player ages their TOI diminishes. 

Percentage of Previous Year’s 5v5 TOI/60 Carrying to Next Year’s (y), Versus Age (x) - Forwards

So, what you see here is basically a multiplier, where from the transitions from 18 to 19 through to the transition from 39 to 40 are expressed as percentages of TOI retained. Meaning, from age 18 to 19 a forward’s TOI/60 typically rose about 10% (retention of 110% of previous year’s TOI/60, presumably because that’s what the young player told his coach he was going to give), and from 39 to 40 the player retained about 95% of the previous year’s TOI/60. Given the strength of fit that this model gives me for the predictive equation (visible in the above chart), I could just roll with that, but I’d probably also want to establish a “saturation point” – in other words, the high-water point where you shouldn’t expect TOI/60 to increase. 

Identifying the “saturation point”; wherein, the highest point of 5v5 TOI/60 before you can expect the following year’s 5v5 TOI/60 to be less…in this case, among forwards.

I won’t set my saturation point at 13m/60, for the simple fact that you can see the data still can easily spike back up to 0 at that point (meaning, the variance can still pull the following year’s TOI/60 to zero change). It appears that 15m/60 is our true point of no return. Also of note: your goons are clearly defined in that crowd below 7m/60, as the group that will tend to continue getting very little 5v5 time. In sum for 5v5 TOI/60, expect your minutes to increase around 5-10% per year up to about 24 (keeping in mind not to expect higher than that saturation point), 2.5% steady up to 29, and hold until around 35 before starting to descend. Your average rookie forward is 22.6 years of age, and begins at around 10.15 5v5 TOI/60, which would make their predictive career arc like so:

Average Forward 5v5 TOI/60 Career Arc, ages 22 to 35. The average rookie forward age is 22.6, and most forwards are no longer in the league after age 35 due to decline and 35+ contract rules.

The arc for forward 5v4 TOI/60 is very interesting compared to 5v5; generally speaking, growth is not expected after 23 years old.

I’ve included a dotted polynomial trendline that had a better fit to the data (r-squared 0.80) but I really can’t intuitively bring myself to project zero change in TOI for forwards entering the league at 18-19 years old. As for our saturation point:

Once again keeping in mind the variance, I’d put it at around 2.75m/60. 

Any fantasy predictor worth its salt takes playing time into account, whether it is considering a player’s line placement, actual time on-ice, or attempting to ground the projections in some hard data. Further refinements that include age trends and saturation points will control and guide the variances that have good explanations. Nailing down these predictors goes a long ways towards building projected boxcar totals, in addition to providing nuance to research on talent and the NHL player population.